Fix sr() on multiple interfaces and libpcap L3 sockets

[gpotter2]

Currently you can send() on multiple interfaces, but not sr().

This PR:

• fixes sr on multiple interfaces by allowing L3socket* to recursively spawn sockets on other interfaces if necessary
• fixes sending on tun-like interfaces through libpcap

[codecov]

Codecov Report

Attention: Patch coverage is 67.39130% with 30 lines in your changes missing coverage. Please review.

Project coverage is 81.67%. Comparing base (420173c) to head (0f68903).
Report is 4 commits behind head on master.

Additional details and impacted files

@@            Coverage Diff             @@
##           master    #4474      +/-   ##
==========================================
- Coverage   81.68%   81.67%   -0.01%     
==========================================
  Files         355      355              
  Lines       84706    84754      +48     
==========================================
+ Hits        69190    69225      +35     
- Misses      15516    15529      +13     

Files	Coverage Δ
scapy/layers/inet6.py	88.47% <100.00%> (+<0.01%)	⬆️
scapy/sendrecv.py	87.21% <100.00%> (ø)
scapy/arch/libpcap.py	76.92% <82.35%> (+0.86%)	⬆️
scapy/arch/linux/__init__.py	83.45% <84.21%> (-0.56%)	⬇️
scapy/arch/bpf/supersocket.py	0.00% <0.00%> (ø)

... and 4 files with indirect coverage changes

[guedou]

Cool PR! I did not test it but it looks good.

[guedou]

Could we use a generic select()method instead of duplicating it in all sockets?

[gpotter2]

I'm not sure if that's clearer, considering the types change in each function.
We would also require a dependency to another L3Socket or something, and I think sockets are already abstracted enough

[guedou]

A classmethod could do the trick. What do you think?

[gpotter2]

sounds like a LOT of refactoring work that I honnestly don't really want to do :(

[guedou]

Your call!

[gpotter2]

You are right though. How about we let it pass today, but I promise I'll come back to it eventually, or whenever we touch it again :)